Associations of air pollution and genetic risk and their interaction with risk of Alzheimer’s disease: identification of risk loci and potential biological pathways

Abstract

Background

Air pollution is associated with Alzheimer’s disease (AD), and the susceptibility of AD from air pollution may be affected by genetic characteristics, but the underlying mechanisms remain unknown.

Methods

Data from the UK Biobank, in conjunction with PLINK2, was utilized to conduct a genome-wide analysis of a gene-air pollution (PM_2.5, PM₁₀, NO₂, and NO_x) interaction in AD. Functional annotation, positional gene mapping, expression Quantitative Trait Loci (eQTL) analysis, 3D chromatin interactions mapping (ciMap), and gene expression analysis were conducted using FUMA. Potential biological pathways was analyzed by Metascape.

Results

A total of 322,958 participants were included in the study. Genome-wide gene-air pollution interaction analysis (GWIA) identified 38, 36, 66, and 101 genomic risk loci that interacted with PM_2.5, PM₁₀, NO₂ and NO_x (p < 5 × 10^–8). The functional annotation positional gene mapping, eQTL, ciMap mapping, and quantitative gene prioritization prioritized 26, 25, 46 and 70 prioritized genes for PM_2.5, PM₁₀, NO₂ and NO_x, including PTCH1, UNC80, TUBGCP3, RUNX2, RCAN2, SUPT3H, DNPH1, TTBK1, and RSPH9, and we found significantly high expression of NO_x-related prioritized genes in the heart atrial appendage, the modulation of smoothened signaling was associated with PM_2.5 and PM₁₀, hemopoiesis was associated with nitrogen oxides, and learning or memory processes may be involved in the interaction between PM_2.5, NO₂, NO_x, and genes, affecting Alzheimer’s disease.

Conclusions

We prioritized genetic risk loci that interact with air pollutants in AD and revealed that learning or memory are crucial pathways through which these pollutants interact with genes PDE1B, SRF, ZNF385A, and TTBK1.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-026-02102-4.

Introduction

Alzheimer’s disease (AD) is one of the most prevalent progressive neurodegenerative diseases. In 2020, about 10% of adults aged 65 and older had AD worldwide, 60–80% of all dementia cases among older adults can be attributed to AD, making AD the most significant cause of dementia [1]. According to the Global Burden of Disease (GBD) Report, the number of people affected by dementia is projected to more than double from 57.4 million in 2019 to 152.8 million cases globally in 2050 [2, 3]. Currently, there is no cure and little in the way of treatments for AD. The Lancet Commission on dementia prevention, intervention, and care highlighted the importance of modifiable risk factors such as air pollution and suggested that up to 40% of dementia prevalence may be avoidable through interventions targeting modifiable risk factors [4]. Therefore, it is crucial to study the effect of modifiable risk factors on AD.

Air pollution poses a significant worldwide risk to human health. As outlined in the GBD report, 4.2 million fatalities are attributed to exposure to fine particulate matter (PM), making it the fifth most significant risk factor for global mortality [5]. Epidemiological literature highlights significant associations between air pollution exposures and AD in humans. A meta-analysis revealed that a 10 μg/m³ increase in PM_2.5was associated with a 4.82 times higher incidence of AD [6]. Research demonstrated that AD patients residing in Kaohsiung, exposed to higher average annual concentrations of NO₂ and PM₁₀, experienced faster cognitive deterioration than Pingtung patients. Higher levels of NO₂ and PM₁₀exposure were associated with an increased risk of AD-related cognitive decline [7]. A European prospective study indicated that participants exposed to NO_x levels of ≥ 50 μg/m³ exhibited a significantly elevated risk for AD. Furthermore, within the high AD genetic risk score group, individuals exposed to NO_x ≥ 50 μg/m³faced a significantly amplified risk of developing AD. The effect of environmental factors on AD may vary, contingent on an individual’s genetic susceptibility or genetic risk factors [8]. Therefore, it is imperative to evaluate further the association between air pollution and genetic susceptibility in AD development to elucidate the effect of air pollution on AD.

AD is acknowledged to have a significant hereditary component [9]. Genetic factors play a role in the occurrence and progression of AD and can also shape the impact of environmental factors on human health. Recent studies have begun to explore genes’ modification role in air pollution’s effect on AD. A population-based birth cohort study found that higher prenatal exposure to coarse particulate matter was differentially associated with greater cerebral white matter volume in APOE ε4 carriers than non-carriers [10]. Furthermore, research indicated that the effect of air pollutants on biomarkers of Alzheimer’s disease varied. For instance, air pollution exposure had a more substantial effect on CSF NfL levels in APOE-ε4carriers. In contrast, the associations between exposure to air pollutants and CL were stronger among non-carriers [11]. Although progressively larger genome-wide association studies (GWAS) have expanded the understanding of AD genetic architecture over the last decade [9, 12], the genetic association between gene-environment interactions and AD remains unclear.

Therefore, this study utilized genome-wide association analysis to explore gene-environment interactions contributing to AD development. Detecting such effects is crucial for understanding the relationship between air pollution, genetic risks, and AD [13]; this knowledge could assist in developing personalized early interventions to reduce the global incidence [13] of AD significantly.

Methods

Study participants

The UK Biobank, a population-based prospective cohort study, recruited over 500,000 residents aged 40–69 in 22 assessment centers in England, Scotland, and Wales from 2006 to 2010. At baseline, each participant’s health and genetic information were collected by completing a series of self-reported questionnaires and genetic, biological, and physical measurements. Each participant provided informed consent before the enrollment, and ethics authorization was approved by the North West Multicenter Research Ethics Committee (REC reference for UK Biobank (11/NW/0382)).

Among the initially recruited 502,411 participants, we first selected a sample of unrelated individuals of predominantly European ancestry, defined by a KING kinship coefficient < 0.0884, ensuring relationships no closer than second degree [14]. Subsequently, we excluded individuals with missing covariate data, missing the first ten principal components of ancestry, and the genotyping array, missing PM_2.5, PM₁₀, NO₂, NO_x data, missing outcomes data, and vascular dementia and the patient had AD diagnosed by a doctor or self-reported at baseline. We obtained a sample of 320,507 individuals (N_case = 1,810; N_control = 318,697). Following this, we randomly divided these individuals, including both cases and controls, into two subsets, with one comprising 70% and the other 30% of the total sample. The former subgroup (70% of the sample) served as the GWAS gene-environment interaction analysis discovery sample. The other subset (30% of the sample) was used to analyze the association between air pollutants and AD within a genetic context (Fig. 1).

Fig. 1.

Analytical framework

Outcome definition

We used the algorithm to identify the primary outcomes of Alzheimer’s disease. The algorithm’s data mainly came from self-reported information, death records, and the International Classification of Diseases code (ICD 9/10) in Electronic Health Record hospitalization. All participants were followed from the date of recruitment until loss to follow-up, death, the first occurrence of Alzheimer’s disease, or the end of this study (depending on the country, September 30, 2021 for England, February 28, 2018 for Wales, and July 31, 2021 for Scotland), whichever occurred first.

Air pollution exposure assessment

Air pollution mainly includes PM_2.5, PM₁₀, NO₂ and NO_x. The exposure data of the above four air pollutants with an accuracy of 1 km × 1 km provided by the UK Department for Environment, Food and Rural Affairs (DEFRA) from 2002 to 2020 were used for this assessment. First, we converted the longitude and latitude of the addresses provided by the participants, and then the bilinear interpolation method was used to match the four spatial cells closest to the address and calculate the corresponding annual average exposure concentration of air pollutants. Finally, the time-weighted method was used to calculate the participants’ exposure to the air pollution. In this study, we evaluated the exposure concentration of each air pollutant from four years before recruitment to the occurrence of AD or the end of follow-up [15].

Genome-wide interaction analysis (GWIS)

We used GWIS to explore the gene-environment interaction between genome-wide Single Nucleotide Polymorphisms (SNPs) and air pollutants in AD. Genotypes collected when participants entered the cohort were used for analysis after genotyping, data imputation, and quality control, which have been described in detail in previous reports [16]. We filtered out all variants with a Minor Allele Frequency (MAF) of less than 0.5%, r² < 0.5 for all SNPs, r² < 0.8 for rare SNPs with MAF < 1%, and variants not conforming to Hardy–Weinberg equilibrium p < 1 × 10^–6, and INFO < 0.3. Quantify the correlation between air pollutants and genetic PC (Table S1). Subsequently, we conducted association analysis using the multiple regression function in PLINK2, adjusting gender, age, the first ten principal components of ancestry, and the genotyping array:

$${\mathit{Phenotype}}_i={\beta }_0+{\beta }_1{\mathit{SNP}}_i+{\beta }_2{A}_i+{\beta }_3{\mathit{SNP}}_i{A}_i+{\mathit{PC}}_{1i}+{\mathit{PC}}_{2i}\cdots ..+{\mathit{PC}}_{10i}+{\mathit{Array}}_i+{\mathit{age}}_i+{\mathit{sex}}_i+{\epsilon }_i$$

where SNP_i is the genotype for individual i, A_i is the pollutant’s concentration, PC1_i…PC_10i represent the values of the principal component, and Array_i is the genotype array (0 and 1 represent the UK Biobank array and the UK BiLEVE array respectively). The P value returned for the β₃ estimates relates to the interaction between SNP and the air pollution value (SNP_iA_i). The regression was performed using PLINK2.

Characteristic genomic loci and Fine-map

The maximum distance between LD blocks to merge into a location should be less than 250 kb to identify the characteristic generic loci. SNP with a p value less than 5 × 10^–8 and r² < 0.6 between SNP sites and the SNP with a smaller p value was defined as independent significant SNPs. To avoid the linkage imbalance effect, we chose the SNP with r² < 0.1 between independent significant SNPs, with the SNP with a smaller p value defined as the lead SNP. Subsequently, Fine-map (https://github.com/finemap/finemap) was used to perform Bayesian fine positioning on the above loci. Under the assumption that “the number of potential causal SNPS ranges from 1 to 10” and “the sum of the posterior probabilities of each SNP is 1”, this method generates a SNP set with a true causal variation probability of at least 95%, providing a core candidate range for subsequent causal variation verification.

Functional annotation

Additionally, candidate SNPs were incorporated in the annotation if they had a maximum p value of 0.05. LocusZoom (https://locuszoom.org) was utilized with 1000 Genomes EUR reference data (Nov 2014 release) to generate regional plots [18].

With the FUMA package (https://fuma.ctglab.nl/) was used for functional annotation and enrichment analysis of GWIS [19], ANNOVAR annotated candidate SNPs. CADD scores elucidate the harmful effect of an SNP, where a CADD > 12.37 suggests potential pathogenicity, with a higher score indicating increased risk [20]. The RegulomeDB score, based on data from eQTLs and chromatin marks, assigns lower scores to SNPs that are more likely to possess regulatory functions. The 15-core chromatin state denotes the accessibility of genomic regions (every 200 base pairs (bp)), and the lower state indicates higher accessibility. The Probability of Being Loss-of-Function Intolerant score (pLI) was used to measure the heritability of SNPS for complex traits enriched near selective constraint genes, with genes exhibiting high-pLI (pLI > 0.9) being reasonable candidates for genes under strong selection [21].

Summary-data-based Mendelian randomization (SMR)

By using aggregated data from genome-wide association studies (GWAS) and expression quantitative trait loci (eQTL) data, the causal association between gene expression levels and complex traits/diseases is revealed, while effectively controlling confounding factors and reducing reverse causal bias. The eQTL data mainly used in this study include: eQTL_Brain_Cerebellar, eQTL_Brain_Cerebellum, eQTL_Brain_Cortex, eQTL_Brain_Frontal, eQTL_Brain_Hippocampus, eQTL_Brain_Hypothalamus, eQTL_Brain_Nucleus, eQTL_Brain_Putamen, eQTL_Brain_Spinal, eQTL_Brain_Substantia, eQTL_Brain_Amygdala, eQTL_Brain_Anterior, eQTL_Brain_Caudate. Calculate the standard error and p-value of the SMR effect size, and conduct the HEIDI (Heterogeneity in Dependent Instruments) test simultaneously to evaluate the validity of the instrumental variables. If HEIDI > 0.05, it indicates no significant heterogeneity, and the instrumental variable hypothesis holds. In this study, an SMR P value < 0.05 and HEIDI Hid > 0.05 were regarded as having a significant and robust causal association between gene expression and complex traits/diseases. Considering the false positive risk brought by multiple tests, the Benjamini-Hochberg (BH) method was used to correct the SMR P values of all detected genes to control the false detection rate (FDR).

Gene mapping

To identify whether any SNP was associated with gene expression, three gene-mapping strategies were used: (1) eQTL Mapping: eQTL mapping maps SNPs by searching databases such as GTEx/v8, eQTLcatalogue, DICE, scRNA_eQTLs, eQTLGen, and cis-eQTLs. Given that AD is characterized by neurodegeneration, brain-specific QTL annotation was performed using databases such as BRAINEAC [22], the CommonMind CONSORTIUM [23] Portal, and xQTL Serve [24]. (2) Positional mapping: This approach is based on physical distance, with a posMapWindowSize = 10 kb. (3) 3D Chromatin Interactions Mapping (ciMap): ciMap maps SNP to genes without distance boundaries. It identifies SNP-gene associations through 3D DNA-DNA interaction, with a ciMapPromWindow set at 250–500 kb and FDR < 1 × 10^–6. First, PsychENCODE, FANTOM EP, and Hic databases were searched for built-in chromatin interactions. Subsequently, we annotated enhancer/promoter regions by searching 111 epigenome databases, including those specific to the brain.

DEG and gene expression analysis

Using the GENE2FUNC tool of FUMA8 [19], GTEx v8 RNA seq data was first used for tissue-specific gene expression heatmap analysis. Subsequently, BrainSpan RNA seq data was utilized to generate a gene expression heatmap and enrichment analysis of DEG. Two-sided Student t-tests were conducted to assess the significance of genes across different tissue types or developmental stages. A gene with a Bonferroni corrected p-value of < 0.05 and a logarithmic multiple change of ≥ 0.58 was defined as DEG.

Metascape analysis

Metascape [25] (https://metascape.org/) was employed for pathway enrichment and protein–protein interaction analysis. Pathway and process enrichment analysis primarily encompassed the KEGG Pathway, the GO Biological process, the Canonical Pathway, Reactome Gene Sets, and Wilk Pathways. The background gene set for enrichment analysis was defined as all protein-coding genes in the human genome (GRCh38 assembly) with valid Entrez Gene IDs (N = 30,286). Multiple-testing correction was performed using the Benjamini–Hochberg procedure to calculate q-values, which control the false discovery rate (FDR) across all tests. Enriched items were filtered using criteria of q-value < 0.01 (after FDR correction), gene count ≥ 3, and enrichment factor > 1.5, followed by clustering of functionally similar terms. Pathway enrichment analyses were performed independently for the prioritized gene lists associated with each pollutant. Additionally, the uniform conversion of all input identifiers to human Entrez Gene IDs via Metascape’s ortholog mapping pipeline ensured consistent gene annotation across pollutant-related gene lists, reducing batch effects from identifier heterogeneity. This approach minimizes the impact of locus overlap on the interpretation of pathway enrichment specific to each pollutant.

Analysis of the association between air pollutants and AD

Firstly, the Cox proportional hazards regression model was used to evaluate the associations between PM_2.5, PM₁₀, NO₂, NO_x and AD. The main covariates adjusted included age, sex (Female, Male), education level (Any school degree, Vocational qualification, College education and Other), BMI, deprivation index, alcohol consumption (Occasional, Moderate, Heavy and Never), physical activity (Lower, Moderate, High), income level (Less than 18,000, 18,000 to 30,999, 31,000 to 51,999, 52,000 to 100,000, Greater than 100,000, Unknow), Healthy diet level (Healthy, Unhealthy). The "rms" package was used to draw restricted cubic spline curves to assess potential nonlinear associations. Subsequently, the quantile-based g-computation method was used to evaluate the comprehensive impact of mixed exposure to multiple pollutants.

In addition, we grouped the lead SNP_S in the air pollution genome-wide interaction analysis according to their genotypes (0/1/2). We used Cox proportional risk models to explore the association between air pollutants and AD. We used the sex, age, annual household income, body mass index, alcohol consumption, physical activity, education level, the first ten principal ancestry components, and genotyping array as covariates.

Sensitivity analysis

The sensitivity analysis is as follows: 1) The association between air pollution and AD was explored by defining Alzheimer’s disease phenotypes using hospitalization and death registration codes. 2) The association between air pollution and AD was analyzed after adjusting for factors such as age, gender, education level, BMI, deprivation index, alcohol consumption, physical activity, and income status using the age of onset model and the death competitive risk model. 3) We conducted association analyses using exposure lag levels of 5 years and 10 years respectively. 4) We used time-varying exposure models to analyze and address issues such as residents’ mobility, and evaluated the association between air pollution and AD. 5) Evaluate the linear relationship between air exposure and PC, assessment center, and Townsend deprivation index (Table S2), analyze the spatial autocorrelation of exposure residuals (Figure S1), and the correlation between exposure and AD PRS (Figure S2). we found residual structure persists. In the GWIS analysis, add adjustments for covariates such as assessment region, deprivation index, smoking, and education, and use environmental five principal components model for analysis. 6) GWIS and gene set analyses were performed when the chr19 APOE region was excluded.

Results

Participant characteristics

A total of 320,507 participants were included in this analysis. Baseline data showed that the average age of the participants was 56.89 ± 7.96 years, among which 53.5% were female. During the subsequent follow-up, 1,810 people developed AD. Compared with the non-affected group, participants diagnosed with AD had a higher proportion of low income and low educational attainment at baseline. In addition, the average (standard deviation) concentrations of PM_2.5, PM₁₀, NO₂ and NO_X in the environment of all participants were 9.78 (1.63) μg/m³, 14.74 (2.28) μg/m³, 17.54 (5.55) μg/m³ and 25.97 (10.09) μg/m³, respectively. Compared with the non-AD group, the concentrations of PM_2.5, PM₁₀, NO₂ and NO_x exposed to AD patients were significantly increased (Table S3).

Associations of air pollutants and AD

Higher levels of PM_2.5, PM₁₀, NO₂ and NO_x were associated with increased incidence of AD (Figure S3), With each increase in IQR exposure, the HRs (95% CI) of 1.599 (1.387,1.843) and 1.491 (1.294, 1.718), 1.270 (1.123, 1.435), 1.235 (1.102, 1.384) respectively in the Cox analysis (Table S4). Quantile-based g-computation analysis was used to estimate the joint effect of mixture of PM_2.5, PM₁₀, NO₂ and NO_x on risk of AD; it was found that the total mixture effect was 1.237 (1.123, 1.363). In addition, we found that mixed exposure to different air pollutants also increased the risk of AD (Table S4). The results of the time-dependent Cox model (Table S5), Fine Grey competitive risk model (Table S5), and age-onset model were stable (Table S5). In the AD population defined by hospitalization and death registration codes also were similar (Table S6). Moreover, the sensitivity analysis showed 5-year lag and 10-year lag in exposure (Table S7). With each increase in IQR exposure of PM_2.5, PM₁₀ and NO₂, the risk of AD increases by 0.257 times, 1.459 times and 1.142 times in 5-year lag exposure respectively, and the results remain relatively stable.

Genome-wide interaction analysis

The results of GWIS analysis are shown in Fig. 2 and Figure S4. We observed significant genome-wide interactions (p < 5 × 10^–8) between PM_2.5, PM₁₀, NO₂, and NO_x relative to AD. Specifically, 178, 154, 268, and 465 signals achieved genome-wide statistical differences, respectively (Table S8-Table S11). Sensitivity analysis showed that after further adjusting for covariates such as the assessment region, Townsend deprivation index, smoking, and education (Table S12-Table S15), or add environmental five principal components (Table S16-Table S19), the results remained stable.

Fig. 2.

Manhattan plots for the genome-wide gene-air pollutant interaction GWIS. (a) Alzheimer-PM_2.5, (b) Alzheimer-PM₁₀, (c) Alzheimer-NO₂, (d) Alzheimer-NO_x. The red line represents a P- value threshold of 5 × 10^–8, red point represents independent significant SNPs

Additionally, we prioritized 43, 40, 77, and 122 lead SNPs from 46, 44, 86, and 142 independent significant SNPs, respectively, on 38, 36, 66, and 101 genomic risk loci (Table S20 and Figure S5). Fine-map further refined the genomic loci and revealed that in the reliable SNP sets of PM_2.5, PM₁₀, NO₂ and NO_x, there were at least one potential causal SNP at loci 29,28, 44 and 70 of the above loci respectively (Table S21- Table S24). Several SNPs, such as rs79642285 (near RP5-1018K9.1), rs11902061(near MAP2), rs10172136 (in UNC80), rs6826908 (in PPP2R2C), rs16897942 (near CTC-537E7.2), rs73428646 (in RSPH9), rs7759853 (in RUNX2), rs113853809 (in RP11-554D15.1), rs80005301 (near RP4-668E10.2), rs1232300 (in RCN1), rs150735818 (near LINC01066), rs5802736(in LINC00457, RP11-16D22.2 and RP11-266E6.3), rs9603907 (in TUBGCP3), rs4575441 (near RP11-454K7.2) and rs35685778 (in RP11-219B17.1), were associated with PM_2.5, PM₁₀, NO₂ and NO_x. Sensitivity analysis revealed that even after excluding the APOE locus, these lead SNPs were still detected (Table S25).

To account for the significant interaction effect, we propose an association between exposures and AD, stratified by genotype (number of allele copies). The allele frequencies of leadSNP in the exposure quantiles were relatively stable (Table S26), we found when the copy number of rs5802736 was 1, the HR (95% CI) between PM_2.5, PM₁₀, NO₂, and NO_X and AD was 1.79 (1.10, 2.93), 1.53 (1.08, 2.17), 1.13 (0.98, 1.30) and 1.06 (0.98, 1.15), respectively, the copy of rs5802736 was 2, the association between PM_2.5, PM₁₀, NO₂ and NO_x and AD was 1.21 (1.14, 1.29), 1.15 (1.08, 1.18), 1.04 (1.02, 1.06) and 1.02 (1.01, 1.03) respectively. Our results indicate a significant association between air pollution and AD in certain instances (Table S27- Table S28).

Functional interpretation of genetic variants

First, all sentinel variants for each locus and variants in LD (r² > 0.6) were annotated with these variants by ANNOVAR. We found that the majority of variants associated with AD were non-coding. Only 0.7% of variants were exonic variants in PM_{2.5(number of variants=2252)} and PM_{10(number of variants=2287)}, 0.5% of 3971 variants were exonic variants in NO₂ and only 0.9% of 6465 variants were exonic variants in NO_x. These exonic variants including rs79642285(in F5), rs11902061(in UNC80), rs73428646(in TTBK1, MRPS18A and XPO5), rs1214750(in ZNF318), rs117695114(in PTCH1), rs74158336(in WDR11), SMUG1, PPP1R1A, rs4575441 (in MIS18BP1), rs56705977(in SLC47A2, ALDH3A1 and ULK2), rs2075650(in TOMM40) and rs6857(in APOE), with potentially harmful effects on gene function. Furthermore, 43.6%, 55.3%, and 48.5% of variants interacting with PM₁₀, NO₂, and NO_x, respectively, were intergenic. Additionally, 34.4% of variants interacting with PM_2.5 were intronic (Fig. 3a). A potentially harmful variant effect (CADD > 12.37) was prioritized for 87,81,123 and 197 loci in PM_2.5, PM₁₀, NO₂, and NO_x, respectively. The highest observed probability of a harmful variant effect was rs4575441 (in MIS18BP1) (CADD = 26.8) in the interaction of four air pollutants. According to the RegulomeDB score, which assigns lower scores to variants with an increased likelihood of having a regulatory function, rs2075650 (in TOMM40) (RDB = 1f) has the lowest scores in PM_2.5 and PM₁₀. Variants such as rs114395485 (in ST8SIA4), rs114739343 (near ZNF318), rs61198573 (in MAD2L1BP), rs16905916 (near RP11-431D12.1), rs370727606 (in PTCH1), rs58475976 (near RN7SKP289), rs115231936 (in RP11-753H16.3:RP11-753H16.5: ZNF385A), rs74049031 (in RP11-16D22.2: LINC00457: RP11-266E6.3), rs76610493 (in ALDH1A2), and rs112323071 (in LARGE) had the lowest scores in NO₂ and NO_x. 15-core chromatin state analysis revealed 160, 151, 204, and 401 loci in PM_2.5, PM₁₀, NO₂ and NO_x, respectively, with the minimum state being 1(Fig. 3b-d).

Fig. 3.

Functional annotation of GWIS results. (a) Functional annotation of statistically significant SNPs using ANNOVR. (b) Distribution of CADD score for variants in genomic risk loci. Score above 12.37 suggests a higher of potential for pathogenicity. (c) Distribution of RegulomeDB score for variants in genomic risk loci, with lower scores indicating a higher likelihood of having a regulatory function. (d) Distribution of minimal chromatin state in 127 different tissues and cells for genomic risk loci on chromosome 6

Candidate gene prioritization at genome-wide loci

We conducted positional, eQTL, and 3D chromatin interactions mapping, which prioritized 364, 356, 546, and 885 unique genes (Figure S6-Figure S9); 105, 80, 121, and 253 of which were implicated by at least two mapping strategies and 12, 12, 28, and 29 genes by all methods. Some loci were found to contain only one prioritized gene. For example, EVA1A (Chr2: 75,696,428–75,796,848), GTDC1 (Chr2: 144,695,635–145,090,135), and VAV2 (Chr9: 136,627,016–136857726) were prioritized for PM₁₀, CCDC93 (Chr2: 118,673,054–118771709) was prioritized for PM_2.5, SP3 (Chr2: 174,771,187- 174,830,430), FABP6 (Chr5: 159,614,374–159,665,742) and ORAOV1 (Chr11: 69,467,844–69,490,184) were prioritized for NO₂, KCNS3 (Chr2: 18,059,114–18542882), SDK1 (Chr7: 3,341,080–4308632) and ATP9B (Chr18: 76,829,285–77,138,283) were prioritized for NO_x. GLIS1 (Chr1: 53,971,910- 54,199,877), TACR1 (Chr2: 75,273,590–75426826), DPP10 (Chr2: 115,199,876–116,603,328), USP37 (Chr2: 219,314,974–219,433,084), CSMD1 (Chr8: 2,792,875–4,852,494), BNIP3L (Chr8: 26,240,414–26363152), FAT3 (Chr11: 92,085,262–92629618), and C14orf166B (Chr14: 77,292,715–77,336,645) were prioritized for NO₂ and NO_x (Figure S10- Figure S13). These genes mentioned above are likely to be the causal genes of AD in that associated region. Other genomic loci included multiple prioritized genes. SMR analysis revealed that for PM_2.5, PM₁₀ and NO_x, PROC (rs2069912) was closely related to Brain Putamen basal ganglia. for PM_2.5, and PM₁₀, NME6 (rs6804986) is associated with Brain Frontal Cortex, for NO₂ and NO_x, RHBDD2 (rs138972761) is related to the Brain Cerebellar Hemisphere. In addition, for PM_2.5, CDC25A (rs11715799) is associated with the Brain Cerebellum, for NO_x, SPOCK1 (rs1528968) is associated with Brain Spinal cord cervical, and LRIT2 (rs4933972) is associated with Brain Caudate basal ganglia (Table S29-Table S32).To evaluate the risk genes for the prioritized genome-wide signals, we used three strategies: (1) annotation and gene-based testing for splicing variants, RDB, CADD, and pLI; (2) the way genes are prioritized; (3) assessment of transcriptomic expression in Alzheimer-relevant tissues [26–29]. We assigned scores according to whether they met the above strategies (Table S33- Table S36). As shown in Fig. 4, PTCH1, UNC80, TUBGCP3, RCAN2, RUNX2, SUPT3H, DNPH1, TTBK1, RSPH9, HOXC6, ZNF385A, CRIP3, and ZNF318 were mapped through three localization methods, RCAN2, DNPH1, WDR11, PTPRC, PLRG1, RSPH9, and ZNF493 are mapped through eQTL association in cerebral cortex, monocytes, and brain caudate basal ganglia, a brain region greatly affected in pathogenesis of AD. Among these genes, PTCH1, UNC80, TUBGCP3, RUNX2, RCAN2, SUPT3H, DNPH1, TTBK1 and RSPH9 were associated with all air pollutants, HOXC6 and ZNF385A were associated with PM_2.5, NO₂, NO_x, CRIP3, ZNF318 and ARHGAP32 were associated with PM₁₀, NO₂, NO_x. Sensitivity analysis revealed that these genes were still mapped when excluding APOE region (Table S37-Table S40).

Fig. 4.

Prioritized Genes from Genome-wide Significant Loci. The priority score is determined by summing evidence scores based in several criteria: (1) Genes containing indSig. SNPs within exon regions receive 1 point. (2) SNPs related to air pollutants with the smallest RegulomeDB score receive 1 point. (3) Genes with the highest CADD score > 12.37 are given 1 point. (4) Genes with pLI > 0.9 are awarded 1 point. (5) Genes located by three independent positional mapping methods are given 1 point. (6) Genes linked to AD-related tissues receive 2 points; others’ tissues receive 1 point. Only genes that can be located by three positioning methods or genes related to at least two air pollutants and with a priority score above 5 are shown

Numerous genes closely associated with AD were prioritized in this genome-wide environment-gene interaction analysis. These genes encompassed factors related to AD, including cerebrospinal fluid t-tau levels, AB1-42 levels, p-tau levels in mild cognit impairment, hippocampal volume, cognitive decline, and verbal memory performance. Notable genes such as PVR, BCAM, TOMM40, APOE, APOC1, APOC4, APOC2, CLPTM1, LASRP, NKPD1, TRAPPC6A, BLOC1S3 were implicated in the analysis of PM_2.5 and PM₁₀. In addition, genes linked to established risk factors like body mass index, type 2 diabetes, and HDL levels for AD demonstrated potential interactions with air pollutants, shaping AD susceptibility (Figure S14A-D).

Tissue-specific expression of interaction signals

We conducted tissue-specific analysis on the preferred genes closely associated with PM_2.5, PM₁₀, NO₂, and NO_x using Gene2function. Leveraging GTE_X v8 RNA-seq data from 54 different tissues, it was observed that some genes had relatively higher expression levels in brain tissues, specifically in caudate basal ganglia, cerebellum, cortex, cerebellar hemisphere, and hippocampus, including MAP2 in PM_2.5, DPP10 and FABP6 in NO₂, ZNF493 in NO_x, CRIP3 in PM₁₀ and NO₂, ARHGAP32 in PM₁₀, NO₂ and NO_x, TTBK1 and UNC80 across all four air pollutants (Fig. 5A). Furthermore_, based on the BrainSpan of developmental stage and age-specific brain expression heat map, we found that compared with other developmental stages of the brain, FAT3, SDK1, SP3, USP37, VAV2, XPO5, ZNF493 exhibited higher expression in the early prenatal and early mid-prenatal stages, the expression of RCAN2, UNC80, ZNF493 and FABP6 showed relatively higher expression levels during early infancy, late infancy, early childhood, late childhood, young adulthood, and middle adulthood. Moreover, MEIS2 was relatively higher during the prenatal stage than subsequent stages (Fig. 5B, C). Differentially expressed gene (DEG) set enrichment analysis of the prioritized genes using GTEx tissues revealed significant differential expression of NO_x-related genes in the heart atrial appendage (Figure S15-Figure S18).

Fig. 5.

Prioritized gene expression analysis for the genome-wide gene-air pollutant interaction. (a) Tissue Expression Heatmaps. Gene expression heatmaps of prioritized genes across all air pollutants, using GTEx V8 (53 tissues). (b) BrainSpan Heatmaps (Different Age Stages). Heatmaps from BrainSpan depicting gene expression across different age stages. (c) BrainSpan Heatmaps (Developmental Stages). Heatmaps from BrainSpan illustrating gene expression across various developmental stages of the brain. Two-sided Student t-tests were performed to assess gene significance across different tissue types or developmental stages. The heatmap displays average expression values (log2 transformed), with darker red indicating higher relative expression compared to darker blue within the same label

Potential biological pathways

We utilized Metascape for analyzing gene networks and protein–protein interactions about prioritized genes (Fig. 6). Our findings suggest an association between learning or memory processes and PM_2.5(log _p-value = −4.05, log _q-value = −0.321), NO₂(log _p-value = −6.77, log _q-value = −2.422), NO₂, and NO_x-related signaling (log _p-value = −6.580, log _q-value = −2.236). Specifically, PM_2.5, NO₂, and NO_x-related genes such as MEIS2, PRKN, PDE1B, SRF, LARGE1, ZNF385A, CSMD1, and TTBK1 are implicated in this process (Table S41- Table S44). The regulation of the smoothened signaling pathway (PM_2.5: log _p-value = −6.084, log _q-value = −1.740; PM₁₀: log _p-value = −6.156, log _q-value = −1.812) may contribute to the effects of particulate matter and assembly of plasma membrane-bounded cell projection (PM_10: log _p-value = −3.394, log _q-value = −0.206) for PM₁₀. Hemopoiesis (NO₂: log _p-value = −5.179, log _q-value = −1.534; NO_x: log _p-value = −4.564, log _q-value = −1.103) may be involved in the effects of nitrogen oxide, respectively. Through protein–protein interaction analysis, interactions between prioritized genes of PM_2.5, such as UNC80 and MAP2 with TTBK1 were found. Similarly, prioritized genes of NO₂, including XPO5 and RUNX2, show interactions with ARHGAP32, ATF7, and SUPT3H, while RORA, SRF, PTCH1, FLI1 and PTPRC exhibit potential interactions. For prioritized genes of NO_x, interactions were observed among XPO5, RUNX2, TGFBR1, SRF, PTCH1, HMGB1, and PTPRC. Potential interactions involving UNC80, DPP10, and TTBK1 concerning nitrogen oxide effects exist.

Fig. 6.

Potential biologic pathways and protein–protein interactions analysis for the genome-wide gene-air pollutant interaction. (a) PM_2.5 related biological pathways (Left) and PPI (Right). (b) PM₁₀ related biological pathways. (c) NO₂ related biological pathways. (Left) and PPI (Right). (d) NO_x related biological pathways. (Left) and PPI (Right). All genes in the genome served as the enrichment background. Items meeting the criteria of p value < 0.01, count >  = 3 and enrichment factor greater than 1.5 were clustered together

Discussion

Based on an extensive, population-based cohort study, we conducted a comprehensive genome-wide analysis of air pollution-gene interactions in Alzheimer’s disease. We prioritized 38, 36, 66, and 101 genomic risk loci that might interact with PM_2.5, PM₁₀, NO₂ and NO_x. Through positional gene mapping, eQTL, ciMap mapping, and quantitative gene prioritization, we found 26, 25, 46, and 70 putative genes for PM_2.5, PM₁₀, NO₂, and NO_x, respectively, including PTCH1, UNC80, TUBGCP3, RUNX2, RCAN2, SUPT3H, DNPH1, TTBK1, and RSPH9, and we found significant high expression of NO_x-related prioritized genes in the heart atrial appendage. Moreover, learning or memory may be involved in the process of PM_2.5, NO₂, and NO_x- gene interaction affecting Alzheimer’s disease, the regulation of the smoothened signaling pathway and assembly of plasma membrane-bounded cell projection assembly may be involved in the effects of particulate matter-gene interactions on Alzheimer’s disease.

We have prioritized PTCH1, UNC80, TUBGCP3, RUNX2, RCAN2, SUPT3H, DNPH1, TTBK1, and RSPH9 in the analysis of four air pollutants, supported by three localization methods. The earliest tau pathology in AD is defined as the accumulation of phosphorylated tau protein (pTau) in the entorhinal cortex (EC). A brain-specific kinase, TTBK1, was found to induce CRMP2 phosphorylation in a transgenic mouse experiment, which led to neuronal mutations and pTau accumulation in EC neurons, thus playing a pivotal role in early AD [30]. Some studies have highlighted the association with AD in Chinese and Spanish populations through specific variants of the TTBK1 gene [30, 31]. Besides gene mutations like amyloid precursor protein, recent research underscores the significance of neurogenesis inhibition and neuronal excitation in neurodegenerative diseases. Inhibition of neurogenesis in the hippocampus and olfactory bulb correlates with decreased olfactory and cognitive abilities in AD [32]. Neurogenesis defects in AD are linked to heightened BMP4 secretion and activation of the Sma, Mad, and RUNX2 signaling pathways. Alleviation of neurogenesis inhibition was observed upon inhibition of the pathway involving RUNX2 [33]. NALCN channels, predominantly expressed in neurons, regulate resting membrane potential and neuronal excitability [34]. UNC80 is an essential component of the NALCN complex, and evidence suggests that mutations in the UNC80 gene may be significantly associated with AD in American families with an onset age between 50 and 60 years, making it a susceptibility locus for AD [35]. Furthermore, research suggests that PTCH1, acting as a receptor, directly interacts with Sonic Hedgehog (SHH), enhancing signal transduction to improve neighboring neuron antioxidant signaling and mitochondrial function, thereby exerting neuroprotective effects in the mature brain [36]. These findings collectively underscore these genes’ pivotal role in AD development. In this study, we also found that air pollutants interact with these genes in affecting AD; this suggests that the above genes are of great significance in the impact of air pollution on AD and deserve further attention.

Despite employing three gene mapping strategies, resulting in multiple putative genes for each GWAS locus, specific genes stood out (EVA1A, GTDC1, VAV2, CCDC93, SP3, FABP6, ORAOV1, KCNS3, SDK1, ATP9B, GLIS1, TACR1, DPP10, USP37, CSMD1, BNIP3L, FAT3 and C14orf166B) as they were the sole genes mapped at their respective locus. Previous research indicates that certain nPM-responsive genes that mediate amyloid clearance vary depending on the APOE allele. In APOE3 carrier women, only VAV2[37, 38] and VAV3 genes may respond to nPM. VAV2 can interact with amyloid precursor protein and regulate the phagocytosis of protofibril Abeta by microglia, suggesting that in the interplay between APOE genotype and air pollution, brain amyloid proteins regulated by VAV2 and others must also be considered. In this study, we also found that VAV2 interacts with PM₁₀at the site Chr9: 136,627,016–136857726 to affect AD occurrence. Additionally, SP3 [39, 40], prioritized in this study, may interact with NO₂ to affect AD, as it is significantly upregulated in expression in AD patients and mouse models. SP3 knockdown can restore synaptic plasticity and memory by significantly improving synaptic function, suggesting its importance in AD. Heather Wood et al. also proposed that blocking HDAC2-SP3 action might be a novel treatment for AD, highlighting SP3’s significance. Furthermore, EVA1A [41], SDK1 [42], GLIS1 [43], TACR1 [44], and BNIP3L [45, 46] have been implicated in AD through GWAS analysis, gene expression, and methylation spectrum analysis, although further research is needed to confirm their roles. Moreover, some genes have not yet been associated with AD. Therefore, these factors still need further research.

Previous studies substantiate the following conclusions: environmental exposure can interact with genes to influence the progression of AD [10, 29]. Our study further identifies certain genes potentially involved in the influence of air pollutants on AD through GWIS analysis. However, their possible biological mechanisms remain elusive. Elucidating the biological mechanisms of putative signals from a functional perspective poses a significant challenge. Employing GO, KEGG, Reactome, and other pathways for functional enrichment, this study unveils biological pathways closely related to the interaction between air pollutants and genes in the AD process. The modulation of smoothened signaling overlapped between differential signals was associated with PM_2.5 and PM₁₀. Regulation of smoothened signaling contributes to development and has a protective effect on adult lifespan and glial protein homeostasis. Activation of smoothened signaling can rescue the harmful effects induced by Aβ1–42 in the AD model [47]. Here, we entified that the prioritized the RUNX2, PTCH1, RORA, and WDR11 genes are involved in smoothened signaling regulation, offering insights into the potential molecular mechanism underlying gene-PM_2.5 and PM₁₀interactions in AD. In addition, our analysis of nitrogen oxides revealed the role of hemopoiesis. Clonal hematopoiesis is characterized by the disproportionate abundance of hematopoietic cell clones harboring specific somatic mutations [48]. In recent studies, increasing epidemiological and experimental evidence indicates that clonal hematopoiesis plays a role in many aging-related diseases [48–50]. Study observed a negative correlation between clonal hematopoiesis-related gene mutations and AD incidence, suggesting a causal relationship between clonal hematopoiesis and AD through Mendelian randomization analysis, revealing that hemopoiesis plays an essential role in AD [51, 52]. This augments our understanding of the mechanism underlying nitrogen oxides and gene interactions influencing AD, as prioritized in our study. The main feature of AD is progressive memory loss. Here, we also unearthed a potential role of learning or memory in the interplay between PDE1B, SRF, ZNF385A, TTBK1 genes, and exposure to the four air pollutants affecting AD. In short, the discovery of these biological pathways guides future functional investigations into the effects of air pollutant exposure on AD and presents promising targets for elucidating the biological explanation of air pollutant exposure in AD.

Our study possesses several strengths. It provides preliminary insights into the association between genetic variants and air pollution on the development and progression of AD through extensive data analyses. We compared our findings with published studies conducted using various methodologies, revealing that the putative risk loci may influence AD development. This suggests their potential involvement in the mechanism of air pollution-gene interactions affecting AD. However, there are many limitations in this study. First, this study was conducted only in individuals of European descent, which still has limitations for extrapolating the effect of the interaction between genes and air pollution on AD. Second, in this study, we wanted to use 30% of the population to verify the effect of each genotype and air pollution interaction on AD under other conditions such as lifestyle and family income, but due to the limited number of people, our verification still cannot involve every genotype found. Third, when conducting exposure assessment in this study, although we used the time-weighted method to control the effects of changes in exposure location and time caused by factors such as moving during the follow-up period to a certain extent, we still lack the effects of indoor air pollution exposure. Fourth, due to potential overlap in genes associated with multiple pollutants, which introduces statistical dependence, direct statistical comparisons of enrichment results (e.g., p-values, odds ratios) between pollutant-specific lists are not performed. Fifth, this study still needs verification from studies of different races and populations. Sixth, our study still lacks animal or cell experiments to further elucidate the mechanism of action of the interaction between gene variation and air pollution on AD.

Conclusion

In conclusion, we prioritized genes that may interact with air pollution to influence AD. Our study provided findings on the association between genetic variants and air pollution on the development and progression of AD, which may help to identify vital genetic subgroups whose risk of AD is more susceptible to air pollution and provide hope for the prevention, development, and implementation of treatment programs for AD.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

Shanshan Ran: Formal analysis, writing – original draft, writing – review & editing. Jingyi Zhang: Formal analysis, writing – review & editing. Miao Cai: writing – review & editing. Lan Chen: writing – review & editing. Zhengmin (Min) Qian: writing – review & editing. Fei Tian: writing – review & editing. Gan Wu: writing – review & editing. Ri’enna Boyd: writing – review & editing. Stephen Edward McMillin: writing – review & editing. Huai-cai Zeng: writing – review & editing. Yin Yang: writing – review & editing. Hualiang Lin: Conceptualization, writing – original draft, writing – review & editing.

Funding

This work is supported by the Open Project Program of the Guangdong Basic and Applied Basic Research Foundation, 2024A1515010465, the Guangxi Key Laboratory of Environmental Exposomics and Entire Lifecycle Health, Guilin Medical University, 2023-GKLEH-01.

Data Availability

All UK Biobank information is available online at the webpage www.ukbiobank.co.uk data are available on application. This research was conducted using the application number 69550.

Declarations

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.